A method for generating optimum decision tree for image retrieval

Abstract

AbstractFor retrieval of image databases there is a method in which image retrieval is based on similarity image input. This retrieval method uses the contents (features) of the input exhibition image as the key for retrieval. However, since much calculation time is required to process the images and to extract features from them, the increase in overall retrieval time poses a problem. Thus, a method is proposed which represents the process of image searching as a decision tree and executes retrieval on the tree. This is an efficient method for reducing the retrieval time since the number of features used for classification therein is small on the average. It also has the merit of decreasing retrieval errors by deleting nodes on which there are many retrieval errors with new nodes with fewer errors. In this paper we consider decision trees for similarity image retrieval, attempt to construct an appropriate decision tree which has a shorter retrieval time and which produces fewer retrieval errors from two given inputs [(1) a set of similarity errors used for retrieval and (2) the retrieval object set of an image database] and perform classification (retrieval) experiments.